Graph-Attention-Based Casual Discovery With Trust Region-Navigated Clipping Policy Optimization.

In many domains of empirical sciences, discovering the causal structure within variables remains an indispensable task. Recently, to tackle unoriented edges or latent assumptions violation suffered by conventional methods, researchers formulated a reinforcement learning (RL) procedure for causal discovery and equipped a REINFORCE algorithm to search for the best rewarded directed acyclic graph. The two keys to the overall performance of the procedure are the robustness of RL methods and the efficient encoding of variables. However, on the one hand, REINFORCE is prone to local convergence and unstable performance during training. Neither trust region policy optimization, being computationally expensive, nor proximal policy optimization (PPO), suffering from aggregate constraint deviation, is a decent alternative for combinatory optimization problems with considerable individual subactions. We propose a trust region-navigated clipping policy optimization method for causal discovery that guarantees both better search efficiency and steadiness in policy optimization, in comparison with REINFORCE, PPO, and our prioritized sampling-guided REINFORCE implementation. On the other hand, to boost the efficient encoding of variables, we propose a refined graph attention encoder called SDGAT that can grasp more feature information without priori neighborhood information. With these improvements, the proposed method outperforms the former RL method in both synthetic and benchmark datasets in terms of output results and optimization robustness.

Active Client Selection for Clustered Federated Learning.

Federated learning (FL) is an emerging distributed machine learning (ML) framework that operates under privacy and communication constraints. To mitigate the data heterogeneity underlying FL, clustered FL (CFL) was proposed to learn customized models for different client groups. However, due to the lack of effective client selection strategies, the CFL process is relatively slow, and the model performance is also limited in the presence of nonindependent and identically distributed (non-IID) client data. In this work, for the first time, we propose selecting participating clients for each cluster with active learning (AL) and call our method active client selection for CFL (ACFL). More specifically, in each ACFL round, each cluster filters out a small set of clients, which are the most informative clients according to some AL metrics e.g., uncertainty sampling, query-by-committee (QBC), loss, and aggregates only its model updates to update the cluster-specific model. We empirically evaluate our ACFL approach on the public MNIST, CIFAR-10, and LEAF synthetic datasets with class-imbalanced settings. Compared with several FL and CFL baselines, the results reveal that ACFL can dramatically speed up the learning process while requiring less client participation and significantly improving model accuracy with a relatively low communication overhead.

Vibration Control of a Constrained Two-Link Flexible Robotic Manipulator With Fixed-Time Convergence.

With the more extensive application of flexible robots, the expectation for flexible manipulators is also increasing rapidly. However, the fast convergence will cause the increase of vibration amplitude to some extent, and it is difficult to obtain vibration suppression and satisfactory transient performance at the same time. In order to deal with the problem, a fixed-time learning control method is proposed to realize the fast convergence. The constraint on system outputs, system uncertainty, and input saturation is addressed under the fixed-time convergence framework. A novel adaptive law for neural networks is integrated into the backstepping method, which enhances the learning rate of neural networks. The imposed constraint on the vibration amplitude is guaranteed by using the barrier Lyapunov function (BLF). Moreover, the chattering problem is addressed by approximating the sign function smoothly. In the end, some simulations have been carried out to show the effectiveness of the proposed method.

Heterogeneity and effectiveness analysis of COVID-19 prevention and control in major cities in China through time-varying reproduction number estimation.

Beginning on December 31, 2019, the large-scale novel coronavirus disease 2019 (COVID-19) emerged in China. Tracking and analysing the heterogeneity and effectiveness of cities' prevention and control of the COVID-19 epidemic is essential to design and adjust epidemic prevention and control measures. The number of newly confirmed cases in 25 of China's most-affected cities for the COVID-19 epidemic from January 11 to February 10 was collected. The heterogeneity and effectiveness of these 25 cities' prevention and control measures for COVID-19 were analysed by using an estimated time-varying reproduction number method and a serial correlation method. The results showed that the effective reproduction number (R) in 25 cities showed a downward trend overall, but there was a significant difference in the R change trends among cities, indicating that there was heterogeneity in the spread and control of COVID-19 in cities. Moreover, the COVID-19 control in 21 of 25 cities was effective, and the risk of infection decreased because their R had dropped below 1 by February 10, 2020. In contrast, the cities of Wuhan, Tianmen, Ezhou and Enshi still had difficulty effectively controlling the COVID-19 epidemic in a short period of time because their R was greater than 1.

A novel framework for the identification of drug target proteins: Combining stacked auto-encoders with a biased support vector machine.

The identification of drug target proteins (IDTP) plays a critical role in biometrics. The aim of this study was to retrieve potential drug target proteins (DTPs) from a collected protein dataset, which represents an overwhelming task of great significance. Previously reported methodologies for this task generally employ protein-protein interactive networks but neglect informative biochemical attributes. We formulated a novel framework utilizing biochemical attributes to address this problem. In the framework, a biased support vector machine (BSVM) was combined with the deep embedded representation extracted using a deep learning model, stacked auto-encoders (SAEs). In cases of non-drug target proteins (NDTPs) contaminated by DTPs, the framework is beneficial due to the efficient representation of the SAE and relief of the imbalance effect by the BSVM. The experimental results demonstrated the effectiveness of our framework, and the generalization capability was confirmed via comparisons to other models. This study is the first to exploit a deep learning model for IDTP. In summary, nearly 23% of the NDTPs were predicted as likely DTPs, which are awaiting further verification based on biomedical experiments.